Connecting to sources…

How to build Roomi open-source cleaning and housekeeping robot: the whole process

10 steps to build Roomi open-source cleaning and housekeeping robot, from buying parts, printing and assembly to software, calibration and training, with the bill of mat…

Not sold and no kit. The creators' estimate: the BOM sheet (2025‑07‑01) totals CA$727 for the steppers, drivers, RP2040 boards, encoders, power supply, two rolls of PLA, aluminium tube, bearings and the storage cart, mostly linked to Amazon; the README's cost table gives US $700, EU $600 and CN $960, all excluding shipping, tools and taxes. The BOM leaves out the on-board Raspberry Pi, the USB-to-TTL adapter for the gripper servo and the modified SO100 used as the leader arm

Buy parts

  • Buy the robot parts

    Buy the parts listed in the Bill of Materials. Cost estimates are US $700, EU $600, and CN $960, excluding shipping, tools, and taxes.

Print and make

  • 3D print the parts

    3D print the Roomi parts.

Assemble

  • Assemble Roomi

    Assemble Roomi using the step-by-step build guide. Assembly is estimated to take about 5 hours.

Software

  • Run the Roomi software

    Flash the firmware, load the code, and run the Roomi software.

  • Install software dependencies

    Install Python 3.8+, Platformio for flashing firmware, and MuJoCo for real or simulated use.

  • Create the Roomi Conda environment

    Create a Conda environment named roomi with Python 3.10, then activate it.

  • Install dependencies and Roomi

    Install the dependencies and install Roomi in editable mode.

  • Register Roomi in ManiSkill

    Open roomi.py to see how Roomi is registered in simulation and how keyframes, sensors, and controllers are defined.

Data and training

  • Run basic MuJoCo teleoperation

    Launch Roomi with MuJoCo and control it through the basic teleoperation script.

  • Run advanced motor control

    Run Roomi using the advanced motor control script.

Parts

  • Actuators: two geared NEMA 23 steppers (wheels), NEMA 17 steppers at 51:1 ×1 and 15:1 ×3 (arm joints), and one STS3215 servo (gripper)
  • Electronics: six TMC2209 stepper drivers, four RP2040 Picos, six AS5600 magnetic encoders, three TCA9548A I²C multiplexers, six limit switches, two 128×64 OLEDs, breadboard and wires, a 24V 20A power supply
  • Structure: two rolls of PLA, an M2–M5 screw set, 200 mm aluminium square tube, 6807 bearings and a storage cart
  • Not in the BOM: the on-board Raspberry Pi (runs rpi_listener.py), a USB-to-TTL adapter for the gripper servo, and the modified SO100 used as the leader arm

Tools

  • A 3D printer (the BOM lists two rolls of PLA)
  • PlatformIO to flash the RP2040 firmware
  • A laptop for the teleoperation script (the keyboard library needs sudo on Linux) and a modified SO100 leader arm

Skills

  • Basic Git, pip and terminal skills (README); Python 3.8+, with a Python 3.10 conda environment for simulation
  • Tuning the closed-loop stepper joints: set target angles and PID gains over serial commands, and home against the limit switches
  • Keyboard control in MuJoCo and registering the Roomi agent in ManiSkill

Software

  • Arm_firmware.cpp: each RP2040 runs two stepper joints closed-loop (TMC2209 + AS5600, PID position control, telemetry every 50 ms)
  • Wheel_firmware.cpp: the two wheel steppers at 1/8 microstepping and a fixed 160 RPM
  • rpi_listener.py (Raspberry Pi): receives UDP commands, forwards them to the RP2040 boards and the gripper servo, and draws the eyes on the two OLEDs
  • teleop.py (laptop): reads the leader-arm servo angles and keys and sends them at 100 Hz
  • roomi_mujoco.py and control.py: keyboard control in MuJoCo; roomi.py: a ManiSkill agent

Documents